Anal. Chem. 1995, 67,4358-4365
On-Line Fluorescence Lifetime Determinations in Capillary Electrophoresis Steven A. Soper,* Benjamin L. Legendre, Jr., and Daryl C. Williams Department of Chemistv, Louisiana State University, Baton Rouge, Louisiana 70803-1804
A near-infrared time-correlated single-photon counting instrument was developed for o n - h e fluorescence lifetime determinations of components separated by capillary electrophoresis (CE). The lifetimes of the migrating components were determined using maximum likelihood estimators,which are computationallyeasy to perform and yield values with high precision and favorable accuracies in the limit of low photocounts within the decay profile. The laser source used in the present system was a passively mode-locked Ti-sapphire laser with a singlephoton avalanche diode serving as the fast detector. The instrument response function of this system was determined to be 165 ps (fwhm). Electrophoretic separation of two near-IR dyes, MTCI (cationic) and IR-125 (anionic), in a 95:5 methanoVwater running b d e r (PH = 9.5) produced electrophoretic peak widths at the base of -1-9 s, which set the integration time for collection of the decay profiles. At a loading level of 1.42 m o l for IR125 and 49 m o l for MTCI, lifetime values were determined to be 482 f 14 ps for IR-125 and 943 & 23 ps for MTCI, which agreed favorably with the lifetimes determined for these dyes using static measurements at high concentrations. To minimize background resulting 6-om scattered photons in ultradilute conditions, which introduces bias into the lifetime determination, the calculation was initiated at a fixed time delay with respect to the excitation pulse. To demonstrate the feasibility of making lifetime determinations in capillary gel electrophoresis, where the gel can produce high scattering backgrounds, the lifetimes of C-terminated fragments produced from the M13mp18 template and labeled at the 5 end of a universal M 1 3 sequencing primer with a nearIR fluorescent tag were determined. The collection of lifetimes for 30 different peaks in the electropherogram yielded a mean value of 581 ps and a standard deviation of 1 9 ps. Several research groups have demonstrated the ability to make on-line fluorescence lifetime (tf)determinations using both and f r e q ~ e n c y ~domain -~ measurements in high-performance liquid chromatographic (HPLC) applications. The advantages (1) Richardson, J. H.; Larson, K. M.; Haugen, G. R.; Johnson, D. C.; Clarkson, J. E. Anal. Chim. Acta 1980,116,407. (2) Imasaka, T.; Kouichi. I.; Ishibashi, N. Anal. Chim. Acta 1982,142,1. (3) Desilets, D. J.; Kissinger. P. T.; Lytle, F. E. Anal. Chem. 1987,59, 1830. (4) Kawabata, Y.; Imasaka. T.; Ishibashi, N. Anal. Chim. Acta 1988,208,255. (5)Cobb, W.T.;McGown, L. B Appl. Spectrosc. 1987,41,1275. (6) Cobb, W. T.; McGown, L. B. Appl. Spectrosc. 1989,43,1363. (7) Cobb, W. T.; McGown. L. B. Anal. Chem. 1990,62,186. (8) Smalley, M. B.; Shaver, J. M.; McGown, L. B. Anal. Chem. 1993,65.3466.
4358 Analytical Chemistry, Vol. 67, No. 23, December 7, 7995
associated with determining lifetimes during an analytical separation include peak identification, determination of coeluting components, and multiplexing applications. Much of the on-line lifetime measurements performed to date have involved the determination of coeluting peaks during the HPLC separation of polyaromatic hydrocarbons @‘AI%). A photophysical characteristic associated with PAHs that simplilies the lifetime determination in HPLC applications is long lifetimes, in the 10-50 ns regime, which relaxes instrumental criteria for making the measurement. McGown and co-workers have used phase-resolved techniques and heterogeneity analysis during the elution process to determine if more than two components were present within the detection zone (coelution) during the chromatographic ~eparation.~-~ By simultaneously monitoring the phase and demodulation lifetimes, the presence of multiple components and their fractional contribution to the decay and chromatographic peak could be determined. Lytle and co-workers have used time-resolved fluorescence techniques to determine coelution during the HPLC separation of PAHs as well.3 In their method, a “ratiogram” was recorded in which the fluorescence was excited with a nanosecond pulsed nitrogen laser and the emission monitored at two different times within the decay. One advantage of using time domain determinations is that time-filtering can be simultaneously implemented, which can improve the signal-to-noise ratio (SNR) during the separation by discriminating against interferences with short lifetimes or scattering photon^.^,^.^ Recently, time-filtering has been used in capillary electrophoresis (CE) with a nanosecond, diode-pumped Nd-YLF laser serving as the excitation source for the fluorescence.l0J1 The authors were able to improve the SNR by discriminating against short-lived interferences present in the biological sample. However, only the time-filtering capabilities of their instrument were used, and measurements of fluorescence liietimes during the CE migration process were not acquired. In applications requiring the monitoring and identification of multiple dyes during the analytical separation, lietimes can improve the efficiency in the identification process when compared to that possible with spectral wavelength discrimination. An example of multifluor analysis is in DNA sequencing, where the identity of a terminal nucleotide base during the separation can be accomplished via differences in spectral emission wavelengths associated with the fluorescent probes.12-1j Identification based on differences in the emission wavelengths can produce errors (9) Haugen, G. R.; Lytle, F. E. Anal. Chem. 1981,53, 1554. (10)Miller, K J.; Lytle, F. E. j . Chromatogr. 1993,648,245. (11) Miller, K. J.; Leesong. I.: Bao, J.; Regnier. F. E.: Lytle, F. E. Anal. Chem. 1993,65,3267. (12) Prober, J. M.: Trainor, G. L.; Dam, R J.; Hobbs, F. W.; Robertson, C. W,; Zagursky, R. J.; Cocuzza, A. J.; Jensen, M. A.; Baumeister, K Science 1987, 238, 336.
0003-2700/95/0367-4358$9.0010 0 1995 American Chemical Society
in the base calling, in some cases due to the broad, overlapping emission profiles which can result in cross-talk between detection channels. Lifetimes can be measured with high precision under appropriate conditions and produce little cross-talk, yielding definitive identification of the particular chromophore during the analytical separation when the dyes have distinct lifetime values. On-line fluorescence lifetime determinations during CE r e p resents new challenges not typically encountered in HPLC applications. These challenges include short residence times of the chromophore within the detection zone and low loading masses. Both of these conditions limit the number of photocounts that can be accumulated into the decay profile, producing poor photon statistics and inaccurate liietimes values with poor precision, especially in highly scattering media. Recently, several groups have demonstrated the ability to determine lifetimes of single molecular events in solution using pulsed-laser excitation and timegated detecti~n.'~-l~ In these studies, decay profiles were constructed from 20-200 photocounts and the lifetimes calculated using simple algorithms to reduce the amount of computational overhead associated with the determination. Several simple computational methods have been developed for calculating fluorescence lifetimes. Two examples include the maximum likelih~od?O-?~ and the rapid lifetime determination method^.?^ While both of these algorithms are computationally less intensive than traditional nonlinear least-squares methods, they can determine only a single lifetime value from a multiexponential decay, whereas nonlinear least-squares will determine the multiple lietimes of the complex decay and the fractional composition of each component. Tellinghuisen and co-workers were able to show that, in the limit of low numbers of photocounts within the decay, the maximum likelihood estimator yielded better precision than least-squares estimator^.'^ In addition, these researchers discovered that background in the form of scattering and impurity fluorescence played a critical role in determining the accuracy and precision of the determination and can increase the error in tfby an order of magnitude unless it could be assessed independently. Near-infrared fluorescence can be an appealing alternative to UV or visible fluorescence when attempting to perform on-line lifetime measurements in ultradilute conditions and complex sample matrices. The advantages of near-IR fluorescence monitoring include smaller Raman cross sections (lower scattering contribution to the background) and fewer fluorescence interference^?^-^* However, a decisive disadvantage associated with near-IR fluorescence is the intrinsically short upper-state (13) Smith. L. M.; Sanders, J. Z.; Kaiser, R. J.; Hughes, P.; Dodd, C.; Connell, C. R; Heiner, C.; Kent, S. B.; Hood, L. E. Nature 1 9 8 6 , 321, 674. (14) Swerdlow, H.; Zhang, J. Z.; Chen, D. Y.; Harke, H. R; Grey, R; Wu, S.; Dovichi, N. J. Anal. Chem. 1991, 63, 2835. (15) Chen, D. Y.; Harke, H. R; Dovichi, N. J. Nucleic Acids Res. 1992,20,4873. (16) Soper, S. A; Davis, L. M.; Shera, E. B. J. Opt. SOC.Am. B 1 9 9 2 , 9, 1761. (17) Wilkerson, C. W.; Goodwin, P. M.; Ambrose, W. P.; Martin, J. C.; Keller, R A Appl. Phys. Lett. 1 9 9 3 , 62, 2030. (18) Tellinghuisen, J.; Wilkerson, C. W. Anal. Chem. 1 9 9 3 , 65, 1240. (19) Tellinghuisen, J.; Goodwin, P. M.: Ambrose, W. P.fMartin, J. C.; Keller, R. A Anal. Chem. 1 9 9 4 , 66, 64. (20) Pierls, R Proc. R. SOC.London 1 9 3 5 , A149, 467. (21) Annis, M.; Cheston, W.; Primakoff, H. Rev. Mod. Phys. 1953, 25, 818. (22) Hall, P.; Selinger, B. J. Phys. Chem. 1 9 8 1 , 85, 2941. (23) Waters, P. D.; Bums, D. H. Appl. Spectrosc. 1993, 47, 111. (24) Ballew, R M.; Demas, J. N. Anal. Chem. 1 9 8 9 , 61, 30. (25) Imasaka. T.; Yoshitake, A; Ishibashi, N. Anal. Chem. 1 9 8 4 , 56, 1077. (26) Sauda, K; Imasaka. T.; Ishibashi, N. Anal. Chem. 1986, 58, 2649. (27) Johnson, P. A; Barber, T. E.; Smith, B. D.; Winefordner, J. D. Anal. Chem. 1 9 8 9 , 61,861.
lifetime associated with the chromophores used in these applications and the poor photophysical properties (Le., low quantum yields, poor photochemical stabilities, and short fluorescence lifetimes) associated with many of these dyes in predominately aqueous solvents.35 Due to the dye's short lifetime, which can range in the 200-1000 ps regime, certain criteria must be associated with the instrumentation in order to perform the analysis using time domain methods. These criteria include subnanosecond pulse widths associated with the excitation source and a detector with a small transit time spread. We have recently shown that highly accurate and precise lifetimes of near-IR fluorescent dyes with subnanosecond lifetimes can be calculated using time-correlated single-photon counting with a passively mode-locked Ti-sapphire laser and a single-photon avalanche diode at low concentrations in static solutions.36 The difficulty associated with the poor photophysics of many near-IR dyes in aqueous solvents can be circumvented to a certain degree by modifying the application to use an organic solvent. For example, we have demonstrated improved sensitivity and resolution in CE applications for the detection of several near-IR fluorescent dyes separated by CE using running buffers composed predominately of methan01.3~The detection limits of some model near-IR dyes separated by CE were found to be in the 100-400 molecules range in running buffers consisting of 95:5 methanol/ water and dropped to 120 000 molecules in 60:40 methanol/water running buffers. The increase in the detection limit with a higher water content was ascribed to quenching effects exhibited by the aqueous solvent on the chromophore. In this work, we investigate the dynamic (on-line) measurement of fluorescence lifetimes for several chromophores separated by free solution capillary electrophoresis using time-correlated single-photon counting in the near-IR The chromophores are tricarbocyanine dyes that show absorption and emission properties in the near-IR and possess lifetimes in predominately methanol solvents which range from -500 to 1000 ps. The instrument used for the present study consisted of a solid-state, passively modelocked Ti-sapphire laser for excitation and a single-photon avalanche diode detector operated in a Geiger mode. The instrument response function of this device has previously been determined to be 165 ps,35,36 appropriate for making subnanosecond lifetime measurements in CE applications. To improve the photophysical characteristics of the dyes during the CE separation, running buffers composed of a high methanol content were used in the separation. In addition, lifetime measurements of DNA sequencing fragments labeled with a near-IR reporter molecule at the 5' end of a primer and separated by capillary gel electrophoresis are presented. The use of lifetime measurements in capillary gel DNA sequencing applications for base calling will be discussed. (28) Imasaka, T.; Ishibashi, N. Anal. Chem. 1990, 62, 363A. (29) Wilberforce, D. -4.; Patonay, G. Spectrochim. Acta 1990, 46A, 1153. (30) Patonay, G.; Antoine, M. D. Anal. Chem. 1991, 63, 321A. (31) Zen, J:M.; Patonay, G . Anal. Chem. 1 9 9 1 , 63, 2934. (32) Williams, R J.; Lipowska, M.; Patonay, G.; Strekowski, L. Anal. Chem. 1993, 65, 601. (33) Soper, S. A; Mattingly, Q . L.; Vegunta, P. Anal. Chem. 1 9 9 3 , 65, 740. (34) Williams, R. J.; Narayanan, N.; Casay, G. A; Lipowska, M.; Strekowski, L.; Patonay, G.; Peralta. J. M.; Tsang, V. C. W. Anal. Chem. 1 9 9 4 , 66, 3102. (35) Soper, S. A; Mattingly, Q . L. J. Am. Chem. SOC.1994, 116, 3744. (36) Soper, S. A.; Legendre, B. L. Appl. Spectrosc. 1994, 48, 400. Legendre, B. L.; Hammer, R. P.; Soper, S. A. Anal. Chem. (37) Flanagan, J. H.; 1995, 67, 341.
Analytical Chemisity, Vol. 67, No. 23, December 1, 1995
4359
-
SPAD Output
MCS
L
Start
r-7-4
Stop
Gate
F
I
1
II
ADC
Figure 1. Near-IR time-correlated single-photon counting device for CE lifetime determinations. L, laser singlet focusing lens; C, capillary tube; BD, beam dump; MO, microscope objective; S, spatial filter; F, bandpass interference filter; SPAD, single-photon avalanche diode; DISC, constant fraction discriminator; MCS, multichannel scaler; TAC, time-to-amplitude converter; ADC, analog-to-digital converter; PC, computer.
EXPERIMENTAL SECTION
Instrumentation. The time-correlated single-photon counting device used for on-line lifetime measurements consisted of a passively mode-locked Ti-sapphire laser (Coherent Lasers, Palo Alto, CA) serving as the excitation source which was pumped by the all-lines output of a small-frame Ar ion laser (see Figure 1). This laser produces subpicosecond pulses at a repetition rate of 76 MHz. The mode structure of the laser is predominately TEMm, which allowed tight focusing of the beam into the separation capillary. The laser light was focused into the capillary using a laser singlet lens (Melles Griot, Irvine, CA) and produced a beam radius of -5 pm (l/e2). The fluorescence emission was collected with a 6Ox high numerical aperture microscope objective (NA = 0.85) obtained from Nikon (Natick, MA). The collected radiation was imaged onto a spatial filter (slit) with a width set at 1.2 mm, giving a viewing distance of 20 pm within the capillary. The collected light was filtered with a Fabrey-Perot interference filter (CWL = 830 nm, HBW = 15 nm, Omega Optical, Brattleboro, VI') and focused onto the face of the photodetector using a 20x microscope objective. The photodetector was a single-photon avalanche diode (SPAD) purchased from EG&G Electrooptics (Vandreille, Canada). The detector was operated at -30 V below its breakdown voltage and possessed a dark count rate of -5075 countds. The output pulses from the SPAD were amplified using a fast amplifier (Phillips Scientific, Mahwah, NJ) and sent into a constant fraction discriminator (CFD, Tennelec Nucleus TC 754, Oak Ridge, TN) for conditioning. The CFD pulses were directed into the gate and stop inputs of a time-to-amplitude converter VAC, Tennelec Nucleus TC863). The TAC was operated in a gated mode of operation such that the conversion rate was set by the photon counting rate. The start pulse for the TAC was generated by an internal photodiode monitoring the laser pulses from the Ti-sapphire laser. These diode pulses were 4360 Analytical Chemistry, Vol. 67, No. 23, December 1, 7995
conditioned by the second channel of the CFD. To preserve the timing, the stop pulses generated from the SPAD were electronically delayed by 13.5 ns prior to the TAC. The decay profiles were constructed by a pulse height analysis (PHA) board and software (PCA II) obtained from Tennelec Nucleus, situated in a PC 486 computer. The output from the TAC was digitized by an AD converter and placed into the appropriate time bin within the decay profile by the PHA hardware and software. The normal intensity electropherogram was constructed by monitoring the fluorescence counting rate during the separation using a multichannel scaler resident in the PC. Free Solution CE. Electrophoresis was performed in a capillary (Polymicro, Phoenix, AZ) with an internal diameter of 75 pm. The optical window was produced using a low-temperature flame to remove the polyimide coating. In free solution CE, the distance from the injection end to detector was set at 45 cm, with a total column length of 50 cm. The high voltage was supplied by a Spellman high-voltage power supply (CZlOOOR, Plainview, NY) and operated in a normal mode (anode at injection end) for free solution CE and a reverse mode for capillary gel electrophoresis. The high-voltage end of the capillary was placed in a protective interlock box constructed in-house. The running buffers used for free solution CE consisted of 95%methanol and 5%triply distilled water. The carrier buffer also contained 20 mM borate with pH set at 9.5. The carrier buffers were not purged with an inert gas to remove 02,since our previous work has shown that 02 has a negligible effect on the photophysics of many tricarbocyanine dyes.38 All near-IR fluorescent dyes were obtained from Kodak (Rochester, NY) and used as received. Stock dye solutions were prepared in 100%methanol at a concentration of 0.1 mM and stored in the dark in a refrigerator at 10 "C. The solutions for CE analysis were prepared daily from serial dilutions of the stock dye at the appropriate concentrations in the running buffer. The column was conditioned daily by passing 1 N NaOH and rinsing with copious amounts of water using a vacuum. The column was then run at 5 kV for 30 min with the running buffer prior to performing the CE analysis. All samples were electrokinetically injected onto the column. Photocounts in the intensity electropherogram were typically integrated for 0.4 s and were not subjected to any type of filtering algorithm prior to presentation. CE Lifetime Determinations. The fluorescence decay profiles were collected over the electrophoretic peak, with data integration commencing when the fluorescence counting rate exceeded the average background rate by -25%, and ceased when the counting rate dropped below this level. The integration time was therefore determined by the time width of the respective electrophoretic band. The lifetimes were calculated using maximum likelihood estimators (MLE) via the following relation,22 m
where m is the number of time bins within the decay profile, Nt is the number of photocounts in the decay spectrum, N; is the number of photocounts in time bin i, and T i s the width of each time bin. A table of values using the left-hand side (LHS) of eq 1 was calculated setting m = 1750 and T = 2.88 ps and using lifetime values (23 ranging from 300 to 1500 ps in 1ps steps. The (38) Soper, S. A.; Huang, J. 1.Photochem., Photobiol., submitted for publication.
start channel for the calculation was shifted, when necessary, by the appropriate number of time bins from the bin with the maximum number of photocounts to reduce the number of scattering photons included into the calculation. The right-hand side (RHS) of eq 1was constructed from the CE decay data over the appropriate range of time bins (5nj. The range of bins used in the calculation was selected to minimize bias introduced by excluding fluorescence photons in the latter time bins and also to reduce the number of background counts included in the calculation at these longer times. The fluorescence lifetime was then determined by matching the value of the RHS obtained from the data with the table entry determined for the LHS of eq 1. The relative standard deviations in the MLE were calculated using,18
-
izooooi 30000
G;
z
10000
o
l 200
.
l
1
400
100 3
3
.
, 600
.
l 800
.
, 1000
,
l 1200
,
, 1400
J L
,
,
1600
,
~
1800
TIME (4
1500
100
IO I
500
To access the bias introduced into the lifetime determination by implementing a time shift in the calculation and to determine the number of scattered photons included in the calculation as a function of the time shift, a Monte Carlo simulation program was developed. The simulated decay profile was constructed by convolving the experimental instrument response function with the exponential decay described by its characteristic lifetime. The appropriate numbers of fluorescence and scattered photocounts were then inserted into the various time bins following a Poisson distribution. Lifetimes were also calculated using the rapid lifetime determination (RLD) method, which is performed by integrating the number of counts within the decay profile over a specified time interval and using the following relationship,
q = -At/ln(D,/DJ
(3)
where At is the time range over which the counts are integrated and DOis the integrated number of counts in the early time interval of the decay spectrum, while D1 represents the integrated number of counts in the later time interval. The relative standard deviation can be calculated from
where ffD, is the standard deviation in DOand U D ~is the standard deviation in D1, determined by taking the square root of the total number of counts in D1 or Do. Preparation of Gel Columns and DNA Sequencing Fragments. The gel columns (polymicro, Phoenix, AZ) were prepared using published proced~res,3~ and therefore, only a brief outline is presented here. The column was cut to a total length of 45 cm, with the distance from injection to detection being 35 cm. The wall of the capillary was preconditioned using the following solutions in the order given: 1 N NaOH (10 min), triply distilled water (10 min), 1N HCl (10 min), and finally triply distilled water (10 min). The wall of the column was derivatized with a 50:50 [3-(methacryloxy)propyl]trimethoxysilane/methanol (Aldrich (39)Williams, D. C.; Soper, S. A. Anal. Chem. 1995, 67, 3427.
1000
TIME CHANNEL(/ 2.88 ps)
1500
Figure 2. Free solution CE electropherogram of DlTCI (A) and IR-125 (B), along with the associated decay profiles. The dye concentration used in both cases was 40.0 pM. The running buffer consisted of 95% methanol and 5% aqueous borate. The dyes were electrokinetically injected onto the column at 5 kV for 5 s, and the separation was performed using a field strength of 491 V/cm. The laser was tuned to 790 nm, with a power of -10 mW at the capillary.
Chemical) solution overnight, followed by drying in an oven at 110 "C. The unpolymerized polyacrylamide gel solution (3%T, 3%C, Sigma Chemicals) was then introduced into the column using aspiration. This gel solution consisted of 8 M urea as the denaturing agent, l x TBE rris-borate, EDTA, pH = 8.3), and riboflavin, which served as the photoinitiator for polymerization. After the column was filled with this gel solution, the capillary was capped at each end, inserted in an ice-water bath, and exposed to UV light for 12 h. Following polymerization, the column ends were trimmed and prerun at 5 kV for a period of 30 min prior to the sequencing run. The sequencing ladder was prepared from the M13mp18 template using standard Sanger dideoxy termination protocols with only the C-terminated fragments generated and the Sequenase Version 2 DNA sequencing kit (United States Biochemical, Cleveland, OH). Two picomoles of a near-IR dye-labeled M13 universal sequencing primer (Li-COR, Lincoln, NE) was annealed to a M13 template at 65 "C for 2 min. The construct was allowed to cool slowly (30 min) to room temperature (25 "C). After cooling, 0.2 pmol of D'IT, 0.1 pmol of MnC12, 2 p L of 1:5 diluted dNTP labeling mix, 0.004 unit of pyrophosphatase, and 0.4 unit of the sequenase enzyme were added to the labeled template. A 3.5 p L aliquot of this mixture was placed in a microcentrifuge tube containing 3.2 pmol of ddCTP. Additional volumes of dNTPs were added to achieve an overall ratio of 1200:l dNTP:ddCTP. This mixture was incubated at 37 "C for a period of 30 min, after which 4.0 p L of a stop solution (95% formamide) was added to the reaction vessel. Prior to injection onto the gel column, the extension mixture was heated to 65 "C and cooled quickly. The sample was electrokinetically injected onto the gel column at 11.25 kV for a period of 2 min. The separation was performed using a field strength of 250 V/cm. RESULTS AND DISCUSSION
Comparison of MLE and RLD Lifetime Algorithms in Free Solution CE. In Figure 2 is shown an intensity electropherogram for the CE separation of the near-IR fluorescent dyes, D'ITCI Analyfical Chemisfy, Vol. 67, No. 23, December 1, 7995
4361
Table 1. Fluorescence Lifetime Determinations in CE Using Maximum Likelihood Estimator (MLE) and Rapid Lifetlme Determination (RLD) Methodss.
dye
cOncn injech 51 @s) g e x p @s) gcalc (Ps) @M) amt (zmol) MLE RLD MLE RLD MLE RLD
DTTCI 400.0 40.0 4.0 IR-125 400.0 40.0 4.0 0.4
4910 491 49 1420 142 14.2 1.42
936 924 943 481 485 483 482
922 939 910 482 487 454 423
7 18 23 4 6 9 14
14 26 33 7 10 11 23
5 7 16 2 2 4 9
8 10 25 2 3 9 12
The experimental standard deviations were determined kom 5-8 replicate measurements. The lifetimes were calculated from the channel with the maximum number of counts over a 5tf time interval, except for the cases when the dye concentration was below 40.0 pM, in which the start time of the calculation was shifted by 30 channels (86 ps) from the channel with the maximum number of photocounts. The calculated standard deviations were determined using eqs 2 and 4. The lietimes for these dyes obtained in static conditions and at high concentrations were 471 f 2 ps for IR-125 and 935 10 ps for D'ITCI.
*
(cationic) and IR-125 (anionic), along with the associated decay profiles. The concentration of the dyes injected onto the CE column in this particular case was 40.0 pM, which represents -491 (D'ITCI) and 142 OR-125) zmol of material. The integration time for construction of these decay profiles was determined by the residence time of the electrophoretic band within the excitation beam and the peak fluorescence intensity and was found to be -2 s for DTTCI and 8 s for IR-125 at this concentration. As can be seen from these decay profiles, DTTCI shows a longer lifetime (smaller slope in the semilog plot) compared to that of IR-125, consistent with the experimentally determined lifetimes of these dyes at high concentrations in static solutions using conventional nonlinear least-squares fitting routines (7f = 471 ps for IR-125; 7f = 935 ps for DTTCD.36 In Table 1are presented lietime data calculated via MLE and RLD using eqs 1 and 3 for several different concentrations of DTTCI and IR-125 separated by CE, along with the standard deviations calculated from replicate measurements (aexp) and using eqs 2 and 4 (uCalc).The time domain over which the calculations were performed at any particular dye concentration was chosen to reduce the number of scattered and background counts included in the calculation. The scattered photons, resulting from Raman and Rayleigh scattered light, can shorten the calculated lifetime because these photons are coincident with the laser pulse. Reduction in the contribution of these photons in the calculation can be accomplished by shifting the start channel in the determination to a latter time bin, due to the temporal characteristics of these photons. A random distribution of background photons in all time bins within the decay profile was also observed. The origin of this background resulted from dark counts associated with the detector and stray light from the capillary tube walls impinging onto the face of the photodetector. The presence of these background photons in the latter time channels biased the calculated liietime to longer values, especially in the lower concentration conditions, where the relative contribution of these photons to the experimental decay was large. To access the relative contributions of scattered and background photons into the calculations at various fluorescence photon levels, Monte Carlo simulations were run with the appropriate lifetime and numbers of scattered, background, and 4362 Analytical Chemistry, Vol. 67, No. 23, December 7, 7995
fluorescence photons to model experimental conditions. When decay profiles were constructed with no background or scattered photocounts and 10 000 fluorescence counts, it was discovered that shifting the start time in the calculation by 28 ps resulted in a lifetime that was -6% longer than the expected lifetime for input values of 471 ps and -3% longer for lifetimes of 950 ps when the MLE method was used. This bias was found to be cumulative up to about four 28 ps time shift intervals. For all MLE determinations, when a time shift was required, this bias was accounted for in the calculated lietimes. When the RLD method was used, no bias was found in the simulation results for a time shift up to 112 ps. The simulation results also indicated that the presence of scattering photons introduced no significant bias when the background-to-fluorescence (B/J?) photocount ratio was < 10%. When the B/F ratio was increased to >20%,signiticant biases were observed. Shifting the start bin of the calculation by 86 ps from the channel with the maximum number of photocounts significantly reduced the scattered photon contribution in the calculation and improved the accuracy. The presence of the background photons into the calculation could be corrected for by subtracting a constant count number from each time bin over the entire decay profile. This constant was determined by averaging the number of counts in each bin from 57r to the last time bin in the decay profile containing counts. As can be seen from the results of Table 1, the lifetimes calculated via MLE at all dye concentrations agreed favorably with the accepted lifetime values for IR-125 and DTTCI. At an injection concentration of 400.0 pM, the fluorescence lifetimes were found to be 481 ps for IR-125 and 936 ps for DTTCI, while at a concentration of 0.4 pM, the CE lifetime for IR-125 was 482 ps. The slightly longer calculated lifetimes for IR-125 in the CE runs resulted from the inability to effectively remove all background photocounts in the latter time bins within the decay profile. For IR-125 at a concentration of 0.4 pM, only 1.42 zmol of material was injected onto the column. Since -10% of the material is effectively sampled due to the size of the separation column (id. = 75 pm), the focused laser beam (w, = 5 pm), and slit width (1.2 mm), the decay profile for IR-125 at this dye concentration was constructed from -85 molecules. This decay profile consisted of -4750 counts and was constructed using an integration time of 1 s in this case. Since the steady-state counting rate in the absence of any dye was determined to be 2700 counts, the net fluorescence counts constituting the decay was -3050 photocounts, with a B/F photocount ratio of 88.5%. Due to the high B/F ratio, a time shift was implemented in this case. When this time shift was inserted (86 ps), the number of counts included into the calculation was 2780. Our simulation results indicated that under these experimental conditions, this time shift reduced the B/F photon ratio to approximately -8.5%. Therefore, only -234 scattered photocounts were included in the calculation under these conditions. In the case of the RLD method, a consistently lower lifetime was found for both dyes at the lower concentrations, even when the appropriate time shift was used. At a concentration of 400.0 pM, the lifetime calculated via RLD for IR-125 was found to be 482 ps, while for DTTCI, the lifetime was 922 ps. When the injected concentration was reduced to 4.0 pM, the lifetimes were determined to be 454 and 910 ps for IR-125 and DTTCI, respectively. These results are consistent with our previous research on static solutions, which indicated that the MLE method
yielded better accuracy in ultradilute conditions, even when a time shift was used.36 The high accuracy obtained in these determinations, especially at the low dye concentrations, results partly from the use of nearIR fluorescence monitoring. Since few compounds show intrinsic fluorescence in the near-IR, negligible interferences are included in the decay profile. This is particularly important when using RLD or MLE algorithms, since they cannot differentiate between a single and multiexponential decays. In the presence of these interferences, the calculated lifetime would represent a weighted average of the various components that constitute the decay, reducing the accuracy in the determination. In addition, the lower Raman scattering cross sections in the near-IR when compared to the visible reduce this contribution in the decay. This then requires shorter time shifts to effectively minimize the scattered photons included into the calculation. Also included in the data of Table 1 are the standard deviations for both MLE and RLD algorithms. The calculated standard deviations were evaluated using eqs 2 and 4, while the experimental standard deviations were obtained from replicate measurements. In general, the MLE method produced smaller calculated and experimental standard deviations, consistent with our previous ultrasensitive lifetime measurements in static solutions.36 For IR-125 at 400.0 pM, the experimental relative standard deviation (RSD) was found to be 0.8% for MLE and 1.5% using IUD, while for D'ITCI at this concentration, the RSDs were 0.7%and 1.5%for MLE and RLD, respectively. At an injection concentration of 4.0 pM, the experimental RSDs were 1.9%and 2.4%for IR-125, and for D'ITCI these deviations were 2.4% and 3.6% for the MLE and RLD methods, respectively. The reduced precision at the lower dye concentrations results from lower numbers of photons included in the calculation and also the increased relative contribution of the scattered and background photons in the determination. Inspection of the standard deviations for the MLE and RLD methods indicated consistently higher experimental standard deviations than calculated standard deviations. The calculated standard deviation represents variability in the determination associated only with the photon statistics (Le., number of counts in the decay profile), while the experimental standard deviation has contributions from both the photon statistics and the experimental conditions, such as instrument stability, dye microenvironment, and scattered and background photon contributions to the decay. In the case of D'ITCI, the dye's photophysics represent a significant contribution to the uncertainty in the measurement. Due to the presence of the two charged alkanesulfonate groups on IR-125, the photophysics of this dye are less dependent on the nature of the solvent, whereas for the singly charged cationic dye, DTTCI, its photophysics are more sensitive to its immediate e n v i r ~ n m e n t . Therefore, ~~ the variability in the lifetimes for D'ITCI from CE run-to-run may result primarily from small changes in the microenvironment producing changes in the dye's lifetimes. In the case of IR-125, the relatively short lifetime of this dye makes it diffkult to effectively minimize the relative contribution of scattered photons in the calculation without sacrificing a large number of fluorescence photons, reducing precision. Peak Identification in Free Solution CE Using Iifetime Matching. To determine the ability to identify unknown peaks in an electropherogram using lifetime matching, a capillary electrophoretic separation of six near-IR dyes was run, with the
C 40000
200
I
1
I
I
I
I
1
400
600
800
1000
TIME (sec) Figure 3. CE separation of six near-IR fluorescent dyes with peak identification via fluorescence lifetime matching. The fluorescence lifetimes were calculated using the MLE method, with the calculation commencing at the channel with the maximum number of counts. All dye concentrationswere set at 100 pM and electrokineticallyinjected onto the column for 5 s at 5 kV, with a separation field strength of 367 V/cm. All other experimental conditions are the same as those used in Figure 2.
lifetimes acquired for each peak in the electropherogram. The results are shown in Figure 3. As can be seen, four of the dyes elute early and are most likely cationic in nature, while the two lateeluting peaks are probably anionic. Decay profiles were constructed over the time width of each band, with the lifetimes calculated via MLE. The identity of each peak was then determined from the calculated lifetime and the known lietime of the components. The assignment of the peaks is given in Figure 3, along with the calculated lifetime values. To confirm the identification, each dye was individually injected onto the CE system and the mobility calculated, from which the migration order was determined. The migration order predicted from lifetime identification agreed with that found from the mobilities of these components. The advantage of lifetime matching for peak identification versus single component injections is that the identification process can be performed in a single CE run, whereas in electrophoretic mobility matching, multiple CE runs are required. Neighboring Peaks and ScatteringInterferences in MLE Lifetime Determinations. To investigate the effects of neighboring components and background on the lifetime calculation using MLE, the integration time for construction of the decay profile was incrementally changed during the electrophoretic separation. Figure 4a shows an expanded view of an electropherogram consisting of three cationic dyes, HITCI (q = 491 ps), IR-140 (732 ps), and IR-132 (697 ps), along with the integration times used in the determination. To allow increasing levels of background and scattered photons into the calculation, the integration time was increased in an electropherogram of IR-140 by itself. Figure 5 shows a series of decay profiles that were constructed from the electropherogram of IR-140 by itself using various integration times. As can be seen from the data of Figure 5, the decay profile collected over a 2 s integration time shows no presence of a prompt peak arising from scattered photons. When the integration time was increased to 25 s, there is evidence of a prompt peak in the early time bins (nonlinear semilog plot), and the background photon level also increased, as shown by Analytical Chemistty, Vol. 67, No. 23, December 1, 1995
4363
I.1 40000
IK.13
IK.140
20000
il
0
:"l
.* $
=
-I+
#
200
220
210
230
Time (s)
,I,
;
......................I . . . . . . . . . . . . . . . . . . .I ...............
. ,
,
,
,
,
,
,
,
080
-I
0 640
0
5
10
15
25
20
Integration Time
35
30
(8)
Figure 4. CE separation of three cationic near-IR dyes (a) and lifetimes as a function of the integration time (concentration,40.0 pM). (b) The integration times and the area they cover during the CE separation are shown in part a. The lifetimes were calculated using MLE with 30 (squares) and 0 channel time shifts (circles). The CE electropherogram consisted of IR-140 only (closed symbols) and HITCI, IR-140, and IR-132 (open symbols). The separation was performed at a field strength of 491 Vicm, with all other conditions similar to those used in obtaining the data of Figure 2. The dashed line shows the fluorescence lifetime of IR-140 calculated at high dye concentrations.
400
0
eo0
2000
1200
I
I
I
!
I
1.00
800
1200
1600
2000
Time Channel (/2.88 ps) Figure 5. Decay profiles for IR-140 only at an injection concentration of 40.0 pM as a function of the integration time over the electropherogram. The dye was injected onto the column for 5 s at 5 kV and electrophoresed using a field strength of 491 V/cm. All other experimental conditions are similar to those used in Figure 2.
increasing number of photocounts in time channels below 400 (prior to excitation) and above channel 1800 (minimal fluorescence photons). Figure 4b shows the calculated lifetime values with and without introducing a time shift (86 ps) into the calculation for IR-140. Increasing the integration time for accumulating counts into the decay for the electropherogram of IR-140 alone indicated that the calculated liietime was independent of the integration time when an 86 ps time shift was used. In the absence of the time shift, a noticeable decrease in the liietime was observed. Since the background is composed primarily of scattered photons, which have a time response coincident with the laser pulse, inclusion of these photons into the calculation would introduce a bias into the determination and lower the calculated value. 4364 Analytical Chemistry, Vol. 67, No. 23, December 1, 1995
When neighboring components are included in the calculation, a significant reduction in the lifetime is observed for IR-140 when a 30 channel time shift (86 ps) is used. The calculated liietime of IR-140 decreased from 728 to 664 ps when the integration included counts from neighboring components. This results from the fact that the calculated value represents a weighted average of the three components that constitute the decay. The reduced lifetime arises from the fact that the two neighboring components have shorter lifetimes than IR-140. When not incorporating a time shift into the determination in the presence of the neighboring components, the liietime does decrease, but not to the same degree as was seen for IR-140 by itself. This results from the fact that more long-lived fluorescence photons are included into the decay with longer integration times than was the case for IR140 by itself. Lifetime Determinations in CGE. To investigate the feasibility of acquiring lifetimes on the fly during the CGE separation of sequencing ladders, C-terminated fragments produced from standard Sanger chain-terminating protocols and labeled with a near-IR fluorophore on the primer were electrophoresed, with lifetimes determined for various components within the electropherogram. The challenge in performing lifetime measurements in CGE when compared to free solution CE results from the gel matrix, which can produce a larger scattering background and introduce impurity components (fluorescence) into the determination. The electropherogram of these C-terminated fragments is shown in Figure 6a. In Figure 6b, an expanded view of a section of the electropherogram is shown, along with the integration periods that were used to construct the decay profiles. The integration time was set at 3-5 s. In Figure 6c,the decay profiles for gel only (prompt peak, no dye-labeled oligonucleotide in detection zone) and the dye-labeled oligonucleotide are shown. When the normalized prompt peaks generated from the gel column were compared to those from the free solution column, no residual fluorescence was observed in the gel case, indicating little fluorescence interference contribution from the gel matrix and other additives when using near-IR excitation. As can be seen from this figure, the choice of the start time in the calculation (86 ps from channel with maximum number of photocounts) eliminated a large fraction of the scattered photons produced by the gel matrix. Using this time shift, concentration, and injection conditions, -20 000 counts were included in the calculation. The average liietime determined using the MLE method was found to be 581 ps, with a standard deviation of f9 ps (RSD = 1.9%). Using eq 2 and the average number of counts included in the calculation produced a standard deviation of approximately f 4 ps. The high precision obtained indicates that minimal fluorescent and scattering interferences contribute to the decay in CGE, a direct result of near-IR fluorescence monitoring. CONCLUSIONS
We have demonstrated the ability to accurately determine the fluorescence lifetimes of components separated via capillary electrophoresis with high precision in the zeptomole regime using time-correlated single photon counting. Two simple algorithms (RLD and MLE) for calculating lifetimes were compared, and the data indicated that MLE produced higher accuracy and precision in ultradilute conditions. Decay profiles for IR-125 were constructed from 85 molecules, and a liietime was determined with high accuracy and precision using MLE during CE. The favorable accuracy and precision was aided by the use of near-IR fluores-
40000
C
I
a
-k
E
51
30000
z
E
2oooo 10000
1800 2000
2200 2400
2600 2800 3000
TIME 25000,
3200
3400
(rec)
1200,
TIME (SEC)
CHANNEL NLrMBER U2.88 ps)
Figure 6. Capillary gel electropherogram of C-terminatedfragments from the M13mp18 plasmid labeled on the 5' end of the sequencing primer with a near-IR fluorescent dye (a) and an expanded view of this electropherogram,showing the time periods (dashed lines) which were used for accumulating photocounts into the decay profile (b). (c) The decay profiles are shown for the dye-labeled oligonucleotide resident in the detection zone (circles) and for the gel only (squares). The laser wavelength was set at 790 nm, with an operating power of 10 mW at the capillary. The sequencing ladder was electrokinetically injected onto the column for 1 min at 200 V/cm, with the separation carried out using the same field strength.
cence monitoring, which minimizes background contributions from scattered and impurity fluorescence photons. Unknown components in CE were definitively identified via lifetime matching in a single electrophoretic run. In addition, the lifetime of C-terminated oligonucleotides separated via CGE were calculated
with high precision. This result indicates that lifetime discrimination can be a viable approach to base-calling in sequencing applications. If definitive identilication of the terminal base could be accomplished with lifetime differences for a series of dyes at 30, then only a 27 ps difference would be required. One disadvantage of near-IR lifetime monitoring is the intrinsic short upper-state lifetimes associated with these dyes, especially in predominately aqueous media. This requires the inclusion of large percentages of organic solvents in the running buffer, such as methanol, which not only lengthens the fluorescence lietime but also improves the signal strength. Even in organic solvents, these dyes display lifetimes in the several hundred picosecond range, requiring a time-correlated instrument with a timing response in the 100-200 ps regime. In the present study, the use of a passively mode-locked Ti-sapphire laser and a singlephoton avalanche diode produced an instrument with the desired characteristics. The avalanche diode is a convenient device for the present application due to its favorable timing response, high sensitivity, low dark count rate, low cost, and long lifetime. While the Ti-sapphire laser is a solid-state instrument, it requires pumping by an Ar ion laser, making it somewhat difficult to operate and cost-prohibitive for the development of a timecorrelated instrument for CE applications. However, pulsed diode lasers can be constructed at a fraction of the cost associated with the Ti-sapphire system and can achieve light levels and pulse widths appropriate for this application. A pulsed-diode laser and avalanche photodiode can be used to construct a low-cost and rugged time-correlated single-photon counting device for liietime measurements in many different analytical applications, such as CE. ACKNOWLEDGMENT The authors thank the National Institutes of Health (First Award), the Louisiana Educational Quality Support Fund, and the Center for Energy Studies 0 for financial support of this work. Received for review June 15, 1995. Accepted September 12, 1995.@ AC950597E @Abstractpublished in Advance ACS Abstracts, October 15, 1995.
Analytical Chertisfry, Vol. 67, No. 23, December 1, 1995
4365
